How Sharksnip Works
A complete walkthrough of the platform — from your first day as a free user to becoming a featured creator earning real money.
TL;DR for the degens
- Free forever to build, train, and backtest models. Live picks are the only paywall.
- $5 Slate Pass for one model + one slate. $10/mo Grinder for everything you build. $30/mo God Mode for active builders.
- Sharp calls = one-click ML helpers (Optimize Weights, Diagnose, Auto-tune). Free tier gets 1/week. Top-up at $1/call any tier.
- Marketplace pays creators flat 50% in cash via Stripe. No tier ladder.
- Consensus = weighted aggregate of every opted-in model. Tip of the iceberg goes to free; full feed is Grinder+.
- We're not a sportsbook and we don't sell signals to books.
Every number on Sharksnip is built on rows like these — 35,000+ NFL plays from 2024 alone, run through the same pipeline that feeds Tinker, Consensus, and the Picks page.
1. Free signup (Sandbox)
Anyone can use Sharksnip free, forever. The free tier (Sandbox) gives you:
- Full Tinker access — train and backtest unlimited models in your browser
- Wire any creator's marketplace blocks into your private models and see the backtest performance (try-before-you-buy)
- The picks page with our predictions across NFL, NBA, MLB, NHL (1-week delay on real-time picks)
- Multi-book pickem unified line shopping (PrizePicks, Underdog, DK Pick 6, Kalshi — pick the cheapest payout)
- A bet tracker that auto-syncs from 30+ sportsbooks
- Free entry into Pack Breaker tournaments
- The top consensus pick per slate (a teaser of the full Consensus product)
- Discord access (general channels)
You'll also get exposure to Shark Snips of the Week, our weekly recap of which marketplace creators ("sharks") hit and missed. That's where most users see the platform's value before paying.
The only thing paywalled is live picks from your trained models. Building, training, backtesting — all free.
2. The four pricing tiers
Sharksnip pricing is simple by design. Pick the row that matches what you actually want:
| You want… | Tier | Price |
|---|---|---|
| To build, backtest, taste 1 Sharp call/week | Free / Sandbox | $0 forever |
| Live picks from one specific model, for one slate | Slate Pass | ~$5 one-off |
| Live picks from all your models + 5 Sharp calls/mo (rollover to 10) | Grinder | $10/mo or $99/yr |
| Everything in Grinder + 50 Sharp calls/mo (rollover to 100) + advanced Consensus | God Mode | $30/mo or $299/yr |
| One additional Sharp call (any tier, any time) | Top-up | $1 per call |
There's a $1 / 7-day Grinder trial if you want to kick the tires. There's also a free 7-day Grinder pass if you sign up for a sportsbook through our affiliate links — that's funded by the books, not by you.
Why Slate Pass exists
Sharp NFL bettors tend to think in slates, not in months. Sometimes you've trained a model and you're 80% sure about it for this Sunday — but you're not ready to commit to a recurring sub. Slate Pass solves this: ~$5, one specific model, one specific slate (e.g., NFL Week 9). Run it as many times as you want during the slate window. No auto-renewal.
Why Grinder is $10/mo
Half the price of the old Pro tier. Cheap enough to be impulse, expensive enough to feel like a real product. Includes live picks from all your trained models, plus a few Sharp calls per month (more on those below).
Why God Mode is $30/mo
For active builders. Same as Grinder for live picks, but with a much larger Sharp call quota (~50/mo) so you can iterate quickly. Adds advanced Consensus features (dispersion, market divergence alerts, custom slicing) and rate-limited API access.
3. Sharp calls — the killer feature
Sharp calls are button-press AI helpers inside the model builder. You don't write a prompt. You click a labeled button. Claude does the ML work behind the scenes and shows you a before/after preview. You accept or reject the change.
The buttons (initial set)
| Button | What it does | Where it lives |
|---|---|---|
| Optimize Weights | Hyperparameter tuning + small architecture search | Training panel |
| Run Monte Carlo | Simulates 10,000 outcomes against historical data; tail-risk summary | Backtesting panel |
| Improve Calibration | Brier-aware probability adjustment (Platt / isotonic) | Evaluation panel |
| Suggest Features | LLM proposes 3-5 new feature engineering ideas | Feature picker |
| Diagnose | Explains why your model is underperforming | Anywhere a model is loaded |
| Auto-tune | Re-trains end-to-end with proposed better hyperparams | Training panel |
How a Sharp call works
- You click the button (e.g., Optimize Weights).
- The system invokes the Claude API with the current model state and a structured tool schema.
- A modal appears showing a headline delta (e.g., "Brier: 0.241 → 0.226"), a what-changed view (architecture diff, feature list, hyperparams), and a projected improvement bound.
- You click Accept (applies the change to your working model) or Reject (closes the modal; quota still consumed).
You're always in the loop. Nothing changes silently.
Quotas
- Free: 1 Sharp call per week (resets every Monday in your local timezone, no rollover)
- Slate Pass alone: 0 Sharp calls (Slate Pass is just a live-picks unlock, not a Sharp call quota)
- Grinder: 5 Sharp calls per month, rolling forward up to 10 if unused
- God Mode: 50 Sharp calls per month, rolling forward up to 100 if unused
- Pack Breaker tournament prizes: can grant bonus Sharp calls on top of your monthly quota
- $1/call top-up: available to every tier as overflow (no subscription required)
Free users get 1 Sharp call per week (resets every Monday). Paid users get monthly quotas that roll over up to 2× their monthly amount — so if you skip a month, your unused calls carry forward, but indefinite hoarding is capped. Anyone — including free users — can buy single Sharp calls for $1 each at any time, no subscription needed. Top-up balance is held separately from your monthly quota and never expires.
Spend order: weekly free call → monthly subscription quota → Pack Breaker bonus balance → $1 top-up balance. This means you'll never accidentally burn a paid top-up while monthly calls are still available.
Monthly quota resets on the 1st of each month, in your local timezone. Weekly free calls reset every Monday. Bonus balance from Pack Breaker rolls forward (capped at 100 to prevent hoarding exploits).
Why this is better than credits
The previous version of Sharksnip had a credit economy ($0.10/credit, credit packs, earnable credits). It worked but required users to translate "30 credits" into "$3" into "is this worth it?" Three mental hops. Sharp calls replace this with: click the button, see the result, accept or reject. Zero hops.
4. The marketplace
This is where Sharksnip gets interesting. Every model in the marketplace was trained by a Sharksnip user (or by us — about 5% of marketplace inventory is "house-published" baseline models).
Buying access to a model
Two options, both running on a clean 50/50 split between creator and Sharksnip:
- Subscribe to the model — recurring monthly at the creator's set price (typical range $5–$25/mo). Unlimited inferences during the sub.
- Slate Pass — one-shot ~$5 to use the model for one slate (e.g., this Sunday's NFL). No auto-renew.
Within your access window, you can run inference on the model as many times as you want. Try it on every game, every prop, every scenario. We deliberately don't charge per-inference because we want you to experiment.
Free users get backtests
A free user can wire any creator's marketplace blocks into their own private model and see the backtest performance. This is genuine try-before-you-buy: you can see how the creator's model would have performed historically, you just can't see this Sunday's live picks until you sub or grab a Slate Pass.
Fork and remix
Click "Fork" on any public model. You get a starter that mirrors the architecture and feature pipeline (NOT the trained weights). Modify, retrain on your own data, publish your version. The original creator is credited on your fork's profile and appears in the public fork tree.
There's no royalty chain — the original creator doesn't earn from your fork's sales. The reward for being forked is visibility and discovery: forks lead users back to the parent, which drives subscriptions to the original.
This is the GitHub model applied to ML — and it's the fastest way to go from "I can't build a model" to "I have a published model" (typically a few hours).
5. Becoming a creator
Anyone can publish a model to the marketplace — Free, Slate Pass, Grinder, or God Mode. There's no paywall on listing.
How earnings work
- You set your model's price (Sharksnip suggests a range; typical $5–$25/mo or ~$5 Slate Pass)
- You keep 50% of every transaction, paid in cash via Stripe Connect
- Sharksnip retains 50% as the platform fee
- Monthly payout cycle, $50 minimum, processed within 7 days of month-end
- 1099-NEC issued at $600+/year (US tax compliance) — handled automatically by Stripe
Featured Creator status (informal, editorial)
Sharksnip's team selects featured creators based on performance, community engagement, and quality. Featured Creators get:
- Prominent placement in marketplace browse
- Featured in Shark Snips of the Week (rotating spotlight)
- Profile badge
- Bonus Sharp calls when featured (typically 25)
Featured status is editorial, not a paid tier. There are no formal Beginner / Established / Signature tiers — that ladder was removed in favor of a flat 50/50 split for everyone, no ladder to climb, no demotion mechanics, no grace periods.
Setting up payouts
On your first sale, the creator dashboard prompts you to set up Stripe Connect (Express account). This takes ~5 minutes and Stripe handles all the W-9 / W-8BEN paperwork. After setup, your monthly payouts arrive automatically.
If you don't want to set up payouts, your earnings accrue until you do. They don't expire.
6. Shark Snips of the Week
Every Monday, we publish a recap of the previous week's marketplace performance. It's our flagship content — and it's what most free users see before becoming paid.
What's in it:
- The top 10 sharks by ROI for the week
- Biggest individual hits (TopShark_Mike's 7-leg parlay that hit 200+ units)
- Biggest misses (with grace, because the failure data is the most educational)
- Contrarian wins (models that diverged from market and hit anyway)
- New featured creators rising (recently editorially selected)
- Strategy spotlight — one creator's methodology featured in detail
- Pack Breaker tournament recap — top tournament submitters + their methodology
We publish it as a blog post, a YouTube video, a Twitter thread, a podcast episode (interview with featured creator), and a TikTok highlight reel. Same data, six channels. Our solo-founder team can sustain this cadence because most of it is auto-generated from platform data.
If your model gets featured, you earn 25 bonus Sharp calls and your creator profile gets a permanent "Featured in Shark Snips" badge.
7. The Consensus product
This is potentially the most valuable thing on the platform. Every model published to the marketplace AND every opted-in Pack Breaker tournament submission contributes to the Sharksnip Consensus — the weighted aggregate of what hundreds (eventually thousands) of independently-trained ML models think.
Every dot is a real model prediction logged against the real outcome. Cherry-pickers don't survive contact with the live track record.
Weighting
Models aren't averaged equally. Better track records earn more weight:
- Verified Sharp models (90+ days, top 10% calibration, 500+ predictions) → 5× weight
- Established models → 2× weight
- Standard models → 1× weight
- Probationary (new) → 0.25× weight (this also applies to Pack Breaker submissions by default)
- Excluded (sub-50% calibration on resolved bets) → 0× weight
These are Consensus contribution tiers — performance grades, not subscription tiers. Don't confuse them with Free / Slate Pass / Grinder / God Mode.
We cap any single creator at 5% of total Consensus weight, so the signal can't be dominated by one popular model.
Forked models get deduplicated
If 30 forks of a top model all vote the same way, that's not 30 votes — it's effectively one. We detect via architecture similarity hash + prediction correlation > 0.95 and treat correlated clusters as combined-weight single votes.
Privacy / opt-in
You only contribute to Consensus if your model is published on the marketplace OR you opt in your Pack Breaker tournament submissions. Private models stay private — train models for your own betting without leaking your edge.
How buyers see it
- Free / Slate Pass: top consensus pick per slate
- Grinder: full Consensus across all games, all sports, dispersion, plus 30-day historical track record
- God Mode: divergence-from-market alerts, custom slicing, full historical, API access
What we won't do
We don't sell Consensus to sportsbooks. Period. Doing so would close the asymmetry between books and bettors on the wrong side, and it would accelerate alpha decay across all our creators. Your data never goes to operators.
8. Pack Breaker (free tournaments)
Sharksnip runs free, skill-based prediction tournaments year-round. No buy-in. No NFTs. No collectibles. Just submit predictions and see how you stack up.
How it works:
- A tournament opens for an upcoming slate (e.g., NFL Week 9 ATS)
- You submit predictions for each game in the slate before the lock time
- Games play out
- The system scores submissions by Brier + ROI, ranks them, and awards prizes:
- Top 1: 1 month of God Mode (or equivalent Sharp call boost) + Snip mention
- Top 10: 5 bonus Sharp calls + a free Slate Pass
- Top 100: 1 free Slate Pass
Prizes are platform credit (bonus Sharp calls + free Slate Passes), never cash, and never tradeable. Pack Breaker is engagement infrastructure, not gambling.
Why it's free: the original plan had paid Pyramid League pack drops (Reignmakers / Sorare-style). We dropped them because (a) at $200+ packs you cross securities-law thresholds and we didn't want that surface area, and (b) the engagement loop works just as well with skill-based prizes funded by platform credit. Tournament submissions also feed into Consensus (with appropriate Probationary weighting), which means free play directly improves the platform signal.
Schedule strength backs every "look-ahead" pick. DET / BAL / BUF aren't winning the implied-total race by accident.
9. Pickem (multi-book line shopping)
If you play DFS-style pickem (PrizePicks, Underdog Pick 6, DraftKings Pick 6), Sharksnip aggregates them into a single interface. You see:
- The same prop across multiple books
- Which book has the best line
- Which book has the best payout multiplier
- Affiliate-tracked direct deeplinks to place the bet
We don't take a cut of your bets — we make money via sportsbook/DFS site affiliate fees when you sign up. The line shopping is free for you and funded by the books.
We've integrated Kalshi (CFTC-regulated prediction market) as a hedge against state-by-state pickem regulations. As regulators continue to crack down on against-the-house pickem, Kalshi's federally-regulated exchange becomes increasingly important.
10. Account, billing, controls
- Cancel anytime — settings page, no friction
- Downgrade anytime — your saved models persist, you just lose Sharp calls and live-picks access until you re-upgrade
- Slate Passes are independent — you can have multiple active Slate Passes (different models, different slates) running alongside any subscription tier (or none)
- Export your data — full export of models, predictions, transaction history
- Privacy controls — granular settings for what's public, what's shared with Consensus, what's available for forking
- Two-factor auth — recommended for any creator accepting cash payouts
What about credits?
Earlier versions of Sharksnip had a credit economy. We removed it. If you have legacy credit balance from before this change, you can spend it on remaining marketplace items that still accept it during the transition window. Going forward, no new credits will be issued.
Questions?
- General: hello@sharksnip.com
- Creator support: creators@sharksnip.com
- Discord: discord.gg/sharksnip
Or just sign up — the free tier covers most of what's described here.